Anomaly detection models assign an anomaly score to an input image between 0
(normal) and 1 (anomalous).
Unlike binary classification models, anomaly detection models are trained using
only positive samples (i.e. "normal" images), instead of the balanced
mix of both positives and negatives you would need to train a classifier well.
This is a self-supervised task.
Datasets follow this structure:
endpoint_url/bucket
├── prefix/images/
└── prefix/metadata.yaml
Dataset images are placed directly inside images/ (subdirectories are ignored).
The metadata file looks like this:
task: anomaly detection